Lab
Small things I built to figure something out.
No roadmap. Sometimes it is a question I wanted answered, sometimes I only wanted to know whether it would work. What stays useful stays up, what gathers dust goes.
Appel & Ei Index
→What one apple and one egg cost in the supermarket, per chain and as a national average.
To see what "for an apple and an egg" is worth these days.
Text as PNG
→Markdown in, dense PNG out, to save tokens for vision models.
Built for one job: the same text for fewer tokens.
unslopf
→A negative-keyword list from a page or a sitemap: recurring filler, not the nouns you have to keep.
Mechanical n-grams. No LLM. Rate-limited API.
Something gets added when I run into something worth building.
For agents
Everything in the lab is callable and readable by machine, with limits.
- MCP server
/mcp— Streamable HTTP, JSON-RPC. Two tools: render_text_as_png and unslopf_negative_keywords.- Lab API description
/lab/api.json— Every endpoint with its parameters, caps, and error codes.- Index for LLMs
/llms.txt,/llms-full.txt— This site as an index, and as one file with the full text.
Every page is also markdown: append .md to the URL. Rate-limited per caller, and switchable off server-side. 429 means slow down, 503 means off.
FAQ
What is the lab?
Small tools that each answer one question: what an apple and an egg cost, whether text as an image is cheaper for a model, which phrases a page already overuses. What stays useful stays here; what gathers dust goes.
Can I call these tools from code?
Yes. Every lab endpoint is described with its parameters, limits, and error codes in /lab/api.json, and the same tools are available as an MCP server at /mcp. Calls are rate-limited per IP.
Do you keep what I send?
No. The lab endpoints do not store the text you submit; they compute and return.